News Samsung announces '10nm-class' LPDDR5

Read more

 

I understand marketing hype as much as (and possibly more) than the next bloke.
"Savings Up to XX%" or "MPG up to", and we all understand, in the round, that it's mainly bs and we're gonna get less.

I also understand v well any £99.99 or £3.99 price structure. My wife still sees those as £99 and £3.

But I simply do not understand why Samsung feel the need to describe "a '10nm-classprocess node' " which as Gareth points out, is "a clever bit of marketing which hides an actual size 'between 10 and 20 nanometres,'"
I'd suggest that virtually everyone, including tech heads, geeks, and nerds, don't really care, so long as it does what it says on the tin.

At NO POINT do I see bragging rights being "I've got 10nm preocess memory in the new trumpet" rather than "I've got DDR5 "

Why bother with that description? Gareth, you're spot on calling them out on it.

 

Unless you're talking about Intel processors

 

which I'm not

as i'm talking DDR5.

If I was Halfacre, I'd quote myself vebatim, but I cant be arsed. I might just get him to do it

 

"Because everyone else does it," basically. Y'see, back in the days of the Intel 4004 you took the smallest feature on the die and that was your process - so, for the 4004, you're looking at 10µm. You kept doing that through 6µm, 3µm, 1.5µm, and so on.

Fast-forward to 65nm, and things get a bit confusing: some "65nm" parts have line widths down to 25nm, while the line pitch can be over 130nm. Gate thickness, meanwhile, is down to 1.2nm, but gate width is 210nm - just 10nm smaller than the 220nm gate widths used in 90nm parts. So, the industry decides that gate length is the important number: a 65nm part should have gates of 65nm or under in length. Job's a good 'un, everyone knows what's what.

This continues as things shrink still further, but hits a roadbump at 22nm when things start getting proper difficult: a typical gate length for a 22nm part would be 25nm - the first time the gate length has been over the "node size". So, back to minimum feature size again: 14nm, then 11nm - which the International Technology Roadmap for Semiconductors measures from the half-pitch of a DRAM part. Only trouble is, 11nm gets thrown out in favour of 10nm - but nothing actually changes in terms of the size. It's purely a branding exercise, pushed further by anything of the "10nm" generation below 20nm being known as "10nm-class".

So, you've got a bunch of manufacturers releasing 10nm and "10nm-class" parts on a node that used to be called 11nm and whose features may be significantly above or below 10nm anyway. For fabs, the number is important both internally and externally: internally it specifies the process node it's using, externally it indicates its process node is competitive with everyone else's process node - so you don't get people taking their orders off GloFo's 13nm process and switching 'em to Samsung's 12nm process when they're actually more or less identical.

For the fans, meanwhile, it's less an actual number and more a generational thing: each node shrink brings with it performance and power advantages, so you want a part built on the latest node. It could have just been generation numbers - we'd be on the 18th full generation if that were the case - but instead it's node size. It doesn't really matter if Samsung's "10nm-class" parts are actually 10nm, it only matters that they're better than the 14nm parts that came before them and competitive with everyone else's "10nm-class" parts.

TL;DR: Welcome to Whose Semiconductor Fab Is It Anyway, where the node sizes are made up and the nanometres don't matter!